Operating Systems and Tweaks Put to Test

[goreblast]

The purpose of this project is two-fold.

Firstly to determine which operating system gives the best results in 1MB, 8MB and 32MB superPI calculations. These three calculations where chosen as they are commonly used in the overclocking/benchmarking community for system speed/comparison/stability testing. SuperPI is a repeatable test, in that it gives very consistent result provided the system remains unchanged.

Secondly to test out the various tweaks that are reported from several internet sources as improving superPI times. Where ever possible each tweak will be tested on each operating system used in the first part of the project.

see full details at http://www.goreblast.com/superPI.htm

[Kasparz]

Bookmarked.

[ZX7891]

how did you not get maxmem to work? it gives a .5s boost

[massman]

does someone know how to see your time with superpi minimised ?

@TS: nice job ! bookmarked !

[bachus_anonym]

@goreblast,

A lot of work done there... Props to you for that :up:

does someone know how to see your time with superpi minimised ?After small window "Calculation Done" pops up:
1. Bring up Task Manager and click on Applications tab,
2. Right-click on "Super PI / mod1.4 and choose "Bring to front",
3. Main Window will show up and you can close Task Manager now ;)

[Fixxxer]

Bookmarked!

Good job! :toast:

[Gautam]

I've never been able to get maxmem to work either. Nice work.

[s7e9h3n]

Thanks for all the hard work :toast: One thing I have to ask though - have you tried to run with a combination of these settings? IIRC, some of the tweaks, when combined with others, have different effects on resulting times. If you were to do that though, I have a feeling it would take you a good part of a couple of months to test all the possible combinations :p:

[zbogorgon]

i tried all tweaks you tried, and some more, and my experiences are quite different.
And IMHO u installed too much drivers and windows updates are really not nessesary.
You should make comparison between, freash install and after installed drives and updates, I allways loose .2s

[goreblast]

Thanks folks.

zbogorgon - I had thought of that, but as this was going to be the first of many tests I thought it prudent to install all CRITICAL updates only. None of the fluff.

I was going to try all the tweaks together after I had finished the round of "Stripping out the rubbish from Windows" but as stated it worked with Windows 2000, but I wrecked the Windows XP instalation and none of the 5 operating systems will boot anymore, and that included Safe Mode.

It was an interesting exercise none the less. And I feel it (maybe not to the same extent) is something that anyone serious about superPI should consider doing. People are always asking which operating system is best for X iteration of SuperPI and what tweaks work. It varies. Test Test Test....

I have to say I was amazed at the gains to be had in 32MB by some of the tweaks, especially in the server and 64bit systems.

[T_M]

Was each test performed with a reboot in between?
Why bother testing closing userinit since it closes itself about 15secs after bootup?
Seems to be a problem with your graphs, the ones after your realtime tests and all those there-after appear to be the same graphs.

Good effort though :)

[G H Z]

I think it would be good to test these from a clean install as well, Slipstream the service packs only. Still this is some very revealing testing that took quite alot of time. I'd be interested in seeing what if any difference there is between system driver version(s), if your going to revisit this that might be a nice addition to your write up.

I'm not sure how the MAXMEM tweak is supposed to work either. When I ran it for PI I used it in combination with Diagnostic Boot and the OS was able to start even with the heavy paging, it does take some time however. You could always try a higher value such as 256 and see if it has any benefit over the normal startup config.

Might be interesting to test if there is any differences in times when deleting all temporary files placed into the PI folder also.

Nice work here gb, thanks for the good info's :toast:

[dinos22]

thanks for the hard work....i've been looking for answers .... :toast:

[goreblast]

T M - I did neglect to change the name of the second column in each graph after the first one, but I assure you they are all different. And they are all in seconds in case that went un noticed.
Userinit.exe doent disappear from those operating systems I tested. I tried it because I have seen it mentioned in various places as being beneficial to superPI times.

I started rebooting before each test, but eventually just defragged, restarted, performed tweak, then tested, performing each test randomly (ie in no particular order). More times than enough the first test carried out was slower than the 2nd or 3rd one at that iteration of superPI.

G H Z - All operating systems where with a clean install and only service packs and windows critical updates. Drivers (except graphics drivers due to the problem that Crossfire drivers do not work in Windows 2000) and those programs mentioned. But I get your point. Ie install windows, install service packs, install superPI, test.

Didnt think of the deleting the temp files from superPI for improving performance. I know you gotta do it to stop errors sometimes.

I might revisit it sometime in the future, but I have got a lot on at the moment so if it does happen it will be a while!

[jrw]

AWSOME! :toast:

[biohead]

O&O Defrag is the best defragmenter. Defragmenting between each install, which doesn't take as long as the first time you defragment, will probably increase performance. With O&O it's best to use /NAME, it will then place everything on the harddrive sorted by name. Personally I experienced a huge performance gain with my normal ATA disk, going from Diskeeper 9 (which I used before) to O&O Defrag. It felt like a cheap raptor.

Good job by the way, you put some time in that. Boring rainy day?

[before]

Very impressive work mate; I've always thought of performing something equivalent myself but never find the necessary patience... :D

[eva2000]

interesting :)

[[cTx]Philosophy]

verry nice work man, im sure this took alot more than one rainy day..
As stated before i lose patience after i install super pi and run 1m run then im off to doing something else, u are a rock man..

[guangg]

nicely done. cheers :D

[K404]

Thanks for putting the work into that!

I found that clearing services and startup under MSConfig gave me up to 0.4 seconds improvement in 1M, and running in high priority helped a small amount too. Both together can shave maybe 0.5 seconds off 1M?

[Clint]

All looks very logically and structurally nicley done...except that last part.

"The above results would show me that for superPI the trimming-down of an operating system by claiming back disk space and removing some eye candy doesn’t benefit the times at all. In fact it might even wreck your otherwise stable system"

..and that is based on the fact that you removed "something" that you don't know what it was... "I gave up and soon discovered that I must have deleted something important!"

....I'm sorry but that kinda ruins the whole documents credibility, better take that part out;)

[sabinus]

Good job goreblast... but there is an error in your graph!
At the section:
Tweak 7– Set Processor Scheduling to ‘programs’ and Memory Usage to ‘system cache’.
it is clear that in Win2003 Server 64bit SuperPI 1MB runs worse with this tweak! You have 32.005 average in the table but you put 31.76 in the graph!!
Judging by the tweaks that GIVE performance boost in SPI 1MB... I thereby come to the conclusion that Win2003 Server is the best.. NOT Win2003 Server 64bit edt.
And you forgot about running each OS with ALL the tweaks that GIVE perf. boost .. That should really conclude the test.